Adaptive Reliability for Optimizing Quality of Information in Sensor Networks
Abstract:
The objective of sensor networks is to deliver actionable information, as opposed to merely connect communication end-points. This makes them mismatched to the design of communication protocols that optimize network-centric goals such as achieving fairness, maximizing throughput, or minimizing delay. The optimization objectives of sensor networks must, in contrast, be information-centric. An example is maximizing the quality of delivered information or minimizing uncertainty in event prediction. A challenge is that the network does not inherently know which bits are more important than others in terms of quality of information. A learning framework is required that feeds back from data fusion results into the operation of key network functions such as reliability and congestion control, to ensure that more resources are allocated to more important information. The components of such a framework are discussed in this talk. Evaluation results demonstrate that the framework can both reduce resources used and improve the information delivered.